Comparison of POISSON - Earlier Version of Excel and Excel 2003

[Jacques25]

Hi everyone,
I have recently started with the statistical functions, which are used in Excel. Now I am studying about the Poisson. I am using it in Excel 2003, since I am having the MS Office 2003. But I want to know about the previous versions of Excel and their differences.!!! So please provide me some useful information about the comparison of POISSON in earlier version of Excel and Excel 2003 Please provide the information as soon as possible.!!

[DANIEL 602]

I am explaining you results about the earlier versions of Excel. The non cumulative Poisson (x, mu, false) uses the following formula :

• EXP (-x) * (mu ^ x) / FACT (x)

Overflow occurs when mu ^ x is too large. This does not occur if mu ^ ^ x 10 ^ 290 (or equivalently x * LOG10 (mu) <290). (FACT x) must also not exceeded. X <= 170 guarantees that. However, earlier versions of Excel does not check these conditions. This usually happens in Microsoft Excel 97.

[Bigga Lexx]

I am telling you about the results in Excel 2003 and later versions of Excel. Because Microsoft has diagnosed when exceeded results by POISSON return the error value # NUM! and knows that POISSON is well-behaved when overflow does not occur, we have added a conditional algorithm in Excel 2003 and later versions of Excel. The algorithm uses POISSON code from earlier versions of Excel, the formula mentioned earlier in this article, when x * LOG10, 290 <(mu) and x <= 170. When x * LOG10 (mu)> = 290 or x> 170, Excel implements a parallel plane described later in this article.

[super soaker]

While talking about the Excel 2003 and later versions of Excel, the secondary plan calculates a sum not scaled probability of each possible observed value. This sum not scaled probabilities is used later for scale purposes. The algorithm also calculates a non-scaled probability that POISSON to return. Finally, it uses the scaling factor to return a correct value POISSON. The algorithm takes advantage of the fact that the relationship between the conditions of the form EXP - x) * (mu ^ x) / x! contains a simple form. The algorithm works as detailed in the pseudocode that is in the following steps. This approach is similar to the method used for the functions BINOMDIST, CRITBINOM, HYPGEOMDIST, and NEGBINOMDIST functions.

[HellBoy21]

Inaccuracies in earlier versions of Excel occur only when mu and x are large enough that x * LOG10 (mu)> = 290 or (FACT x) than for x> 170. In this case, POISSON returns # NUM! In earlier versions of Excel because mu ^ x overflows. To resolve this problem in Excel 2003 and later versions of Excel, you will have to implement the alternative algorithm. nitialize TotalUnscaledProbability and UnscaledResult to 0. Initialize the constant EssentiallyZero to a very small number, eg 10 ^ (-12).

[Zenon]

In earlier versions of Excel functions BINOMDIST, CRITBINOM, function and HYPGEOMDIST NEGBINOMDIST behave the same way. In all cases, these functions also either return correct numeric results or # number! or # DIV / 0!. Again, problems due to overflow or overrun precision. It is easy to identify the conditions where these problems occur, and Excel 2003 and later versions of Excel implement an algorithm secondary plan, such as that POISSON, to return the correct answers in cases where previous versions of 'Excel return # NUM!.